Content uploaded by Hedwig Spegel
Author content
All content in this area was uploaded by Hedwig Spegel on Jan 08, 2021
Content may be subject to copyright.
Effect of classroom air quality on studentsÕconcentration: results of
a cluster-randomized cross-over experimental study
Introduction
In recent years, indoor environments in schools have
come into the focus of discussion. In particular, the
impact of indoor air quality on the attention and CP,
achievements, well-being, and health of students has
been discussed (Daisey et al., 2003; Haverinen-Shaugh-
nessy et al., 2011; Mendell and Heath, 2005; Shendell
et al., 2004).
Carbon dioxide (CO
2
) has been commonly used as
an indicator of indoor air quality. According to The
German Working Group on Indoor Guidelines of the
Federal Environment Agency and the StatesÕHealth
Authorities, air quality can be regarded as ÔharmlessÕif
CO
2
levels are below 1000 ppm, ÔelevatedÕif between
1000 and 2000 ppm, and Ôhygienically unacceptableÕif
above 2000 ppm (Lahrz et al., 2008). This is in line
with guidelines from other European countries
(BMLFUW, 2006; UK Department of Education,
2006; NO-Folkehelseinstituttet 1996).
However, particularly in wintertime, increased CO
2
levels have been observed in classrooms. In a Bavarian
measurement campaign in 91 classrooms, median CO
2
levels ranged between 598 and 4172 ppm (Fromme
et al., 2008). In 25% of the classrooms, the median CO
2
level exceeded 2000 ppm and in 10%, 2700 ppm. Most
Abstract To assess the effect of indoor air quality as indicated by the median
carbon dioxide (CO
2
) level in the classroom on the concentration performance
(CP) of students, a cross-over cluster-randomized experimental study was con-
ducted in 20 classrooms with mechanical ventilation systems. Test conditions
ÔworseÕ(median CO
2
level on average 2115 ppm) and ÔbetterÕ(median CO
2
level
on average 1045 ppm) were established by the regulation of the mechanical
ventilation system on two days in one week each in every classroom. Concen-
tration performance was quantified in students of grade three and four by the
use of the d2-test and its primary parameter ÔCPÕand secondary parameters Ôtotal
number of characters processedÕ(TN) and Ôtotal number of errorsÕ(TE). 2366
d2-tests from 417 students could be used in analysis. In hierarchical linear
regression accounting for repeated measurements, no significant effect of the
experimental condition on CP or TN could be observed. However, TE was
increased significantly by 1.65 (95% confidence interval 0.42–2.87) in ÔworseÕ
compared to ÔbetterÕcondition. Thus, low air quality in classrooms as indicated
by increased CO
2
levels does not reduce overall short-term CP in students, but
appears to increase the error rate.
D. Twardella
1
, W. Matzen
2
,
T. Lahrz
3
, R. Burghardt
3
,
H. Spegel
1
, L. Hendrowarsito
1
,
A. C. Frenzel
4
, H. Fromme
2
1
Department of Occupational and Environmental
Health, Bavarian Health and Food Safety Authority,
Munich,
2
Department of Toxicology and Chemical
Safety, Bavarian Health and Food Safety Authority,
Munich,
3
Berlin-Brandenburg State Laboratory,
Department of Environmental Health Protection, Berlin,
4
Department of Psychology, University of Augsburg,
Augsburg, Germany
Key words: Attention; Carbon Dioxide; Indoor Air;
Students; School.
D. Twardella
Department of Occupational and Environmental Health
Bavarian Health and Food Safety Authority
Pfarrstrasse 3, 80538 Munich
Germany
Tel.: ++49-9131-6808 4249
Fax: ++49-9131-6808 4297
e-mail: dorothee.twardella@lgl.bayern.de
Received for review 26 October 2011. Accepted for
publication 15 February 2012.
Practical Implications
This study could not confirm that low air quality in classrooms as indicated by increased CO
2
levels reduces short-term
concentration performance (CP) in students; however, it appears to affect processing accuracy negatively. To ensure a
high level of accuracy, good air quality characterized, for example, by low CO
2
concentration should be maintained in
classrooms.
Indoor Air 2012; 22: 378–387
wileyonlinelibrary.com/journal/ina
Printed in Singapore. All rights reserved
2012 John Wiley & Sons A/S
INDOOR AIR
doi:10.1111/j.1600-0668.2012.00774.x
378
classrooms rely on natural ventilation, and the cold
temperature of outdoor air inhibits frequent window
opening and causes accumulation of CO
2
in indoor air.
High CO
2
levels in schools have been reported also from
other countries (Daisey et al., 2003).
The relevance of air quality as indicated by CO
2
level
in the classroom for the attention and concentration of
the students could not be consistently shown yet. In a
literature review from 2005, one single publication was
identified, which analyzed the performance depending
on indoor air quality (Mendell and Heath, 2005). A
negative effect of low air quality on reaction time and
performance was reported. More recent publications
provide further indications for an effect of air quality on
attention and concentration (Coley et al., 2007; Ribic,
2008; Wargocki and Wyon, 2007a,b). However, results
are not sufficiently conclusive yet, because multiple
testing as well as lack in blinding may have biased
observed associations. On this background, we initiated
the hereby reported RaBe (Raumluftqualita
¨t und das
Befinden von Kindern – Indoor air quality and student
experiences) study. The objective of the RaBe study was
to assess the effect of indoor air quality as indicated by
the CO
2
level in the classroom on the CP of students.
Methods
We conducted an experimental study with a cluster-
randomized cross-over design. Data collection took
place between November 2009 and April 2010. The
RaBe study was approved by the Ethics Board of
the Bavarian Chamber of Physicians, Germany and by
the Bavarian Ministry of Education and Culture.
Participants
Classes of grade three and four (children usually aged
9–10 years) with a mechanical ventilation system in the
classroom were eligible for participation. Primary
schools in the German states, Bavaria and Berlin/
Brandenburg, were approached by mail as well as
personally, and consent for participation was obtained
from the schools headmasters. Within each class,
parents were informed about the study aims and
procedures by written material as well as by an
information evening in each school, and consent for
data collection was obtained from parents of each
student. All students of the identified classes were
eligible for participation.
We included six schools in our study, five of which
were located in the State of Bavaria and one close to
Berlin. Most of the schools had been recently reno-
vated or built. Per school two to six classes participated
in the study. The classes consisted of 16–29 students of
which between 67% and 100% agreed to participate.
Overall, 417 students from 20 classes took part in the
study (overall response rate for students 84%).
Experimental conditions
In each classroom, three experimental conditions were
implemented:
•ÔUsualÕ: The mechanical ventilation was adjusted as
usual. Window opening was allowed according to
schools general regulations.
•ÔWorseÕ: The mechanical ventilation was down-
regulated. Window opening was not allowed. It was
aimed to reach a median CO
2
level of 2000–
2500 ppm. Thus, conditions as can be found in
wintertime in poorly ventilated classrooms were
simulated.
•ÔBetterÕ: The mechanical ventilation was up-regu-
lated. It was aimed to reach a low median CO
2
level
of <1000 ppm. Window opening was not allowed.
Thus, conditions as recommended by expert panels
were simulated.
The reference period for the experimental condition
was the beginning of the first class hour until the end of
the d2-test in the fourth class hour (not counting
breaks) and thus resulted in a exposure duration of
typically 155 min. Experimental conditions were imple-
mented on 2 days per week. Thus, the data collection
period included 6 days in three consecutive weeks. For
each class, those weekdays were preferably selected as
test days, in which according to the timetable, all
students usually stayed in the classroom all morning (at
least from about 8 oÕclock until 11 oÕclock; i.e., German
class hours 1–4). Classes from the same school were
evaluated parallel during the same 3 weeks. Different
schools were assessed consecutively.
Randomization
We randomized the unit Ôschool class,Õbecause air
quality could only be regulated classroom wise. In all
classrooms, the experimental condition in the first week
(the first two experimental days) was ÔusualÕ. The
sequence in the following 2 weeks – either worse/better
or better/worse – was randomized. If the air quality in
the classrooms within schools could be regulated
separately, classrooms were arranged in pairs and
randomization was performed within these pairs. In
that way, in each pair, one classroom was assigned the
sequence worse/better and the other the sequence
better/worse. Otherwise, the sequence of conditions
was determined for all classrooms within one school
combined. Allocation was performed by random
drawing of marked pieces of paper.
Outcomes
Concentration performance was assessed by the d2-test
(Bates and Lemay, 2004; Brickenkamp, 2002). The d2-
test is a one-page, paper-and-pencil test with 14 rows of
Effect of classroom air quality on studentsÕconcentration
379
the characters ÔdÕand ÔpÕ. The task is to mark as many
target characters as possible (a ÔdÕwith a total of two
dashes placed above and/or below) per row in a limited
time of 20 s. The test person is told verbally every 20 s
to move on to the next line, leaving the previous line
not fully examined. Based on the number of processed
characters and the number of errors, specific outcome
parameters can be determined. Those are Ôtotal number
of characters processedÕ(TN) as an indicator for
processing speed and the Ôtotal errorsÕ(TE, defined as
sum of incorrectly marked distractor characters plus
left-out targets) as an indicator of accuracy, as well as
ÔCPÕ(defined as the number of correctly marked target
characters minus incorrectly marked distractor char-
acters) as an indicator for overall concentration. In
RaBe, CP was used as a primary outcome. The d2-test
was applied at each of the six experimental days in the
fourth class hour. Trained study personnel instructed
the students on how to fill in the form according to
standard instruction. The students, teachers, and
instructors were blinded with respect to the experimen-
tal condition on the day of the test. Non-participating
children stayed in the classroom, and where given
different tasks during the time, participating children
filled out the questionnaire.
Other data collection
Sociodemographic data on the participating children
were collected via a standardized questionnaire for
parents. Characteristics of the schools and the class-
rooms were collected by a standardized documentation
sheet, which was filled out by study assistants.
Measurement of air quality
During the 3 weeks of data collection, we documented
the air quality in the respective classrooms. For this
purpose, we placed an air-quality sensor (Klimawa
¨ch-
ter MF420-IR-CTF; J. Dittrich Elektronic GmbH &
Co KG, Baden-Baden, Germany) in the middle of the
classroom above the heads of the children. For each
minute, the average room temperature (measurement
range, 0–50C), relative humidity (measurement range,
15–95%), and CO
2
level (measurement range, 0–
3000 ppm) were recorded. The median CO
2
level,
temperature, and relative humidity in a classroom on
an experimental day were determined from the minute-
by-minute measurements during the first four school
hours (not counting the breaks) until the end of the d2-
test. In case of gaps in data of maximal length of ten
minutes (seven occasions), missing data were substi-
tuted by the average of the adjacent values. In case of
longer gaps (nine occasions), missing values were
substituted by measurement results from the second
day with the same experimental condition in the
respective classroom.
Statistical methods
We compared the distribution of baseline characteris-
tics of classes and students in the two experimental
arms (sequence usual–worse–better vs. sequence usual–
better–worse). Grade and sex of the students were
taken from school lists, other sociodemographic data
from the parental questionnaire. We determined rela-
tive poverty based on the equivalent household net
income, which was calculated from the household net
income and the number and age of the household
members weighted according to OECD (Bundesregie-
rung 2005). If the equivalent household net income
was <60% of the median equivalent net income in
Germany, relative poverty was categorized as ÔyesÕ.
Next, we described the CO
2
level in the classrooms
depending on the experimental condition as well as the
distribution of the parameters of the d2-test.
To determine the effect of the experimental condition
on the test results (CP, TN, and TE), we employed
three-level hierarchical linear models (Raudenbush and
Bryk, 2002). Hierarchical linear models are an exten-
sion of the linear model, which is used in RaBe to
account for the repeated measurements of students as
well as for the correlation between students belonging
to the same class. By defining test parameters on level
one, students on level two, and classes on level three,
the measured test parameters were examined within
students within classes. The model assumes normally
distributed errors on all three levels. Results of analysis
are given on an additive scale.
First, we set up to a growth model to model the
change in parameters with the repeated application of
the d2-test (i.e., the learning effect). We allowed a linear
and a quadratic term of growth. The inclusion of these
terms as fixed effects on level one as well as random
effects on level two and three was tested for significance
at P< 0.05 based on the likelihood ratio test. In
the resulting growth model, parameters to assess the
impact of air quality were included as follows:
We followed three different approaches to test our
hypothesis of air-quality effects on student perfor-
mance in the d2. First, we used experimental condition
(usual/worse/better) coded as dummy variables as
predictors of the CP and included the complete data
set. This analysis is in accordance with the design and
comparable to an intention-to-treat approach in clin-
ical studies. In an intention-to-treat approach, the
complete data set has to be included in analysis and
participants have to be analyzed as randomized irre-
spective of whether the treatment was in fact imple-
mented or not.
As in some occasions the study protocol was not
followed exactly, analysis was repeated using a reduced
data set after exclusion of observations with deviations.
Deviations from study protocol pertained to the
completion of the d2-test (30 s instead of 20 s given
Twardella et al.
380
for the completion of one of the 14 lines in the d2-test,
student did not use the prescribed sign to mark relevant
letters, student did not complete all 14 lines) as well as
to test conditions (the whole class left the classroom for
at least one hour during the morning, a part of the class
left the classroom for at least one hour during morning,
the class had physical education during the morning,
the regulation of the ventilation system did not work).
If the deviations from the study protocol were signif-
icantly associated with the test parameters in the
hierarchical model and thus could introduce confound-
ing, observations were excluded.
Finally, because of the variability of actual CO
2
levels within the study groups, in a third analysis, the
association of the actual CO
2
median with studentsÕ
concentration was estimated. In this analysis, instead
of the study condition, the median CO
2
levels were
introduced as a linear predictor in the model, as
ventilation rate was found to be linearly related to
student achievement (Haverinen-Shaughnessy et al.,
2011). Analysis was based on the reduced data set as
in the second analysis.
Below, we present results from three analyses:
analysis 1 with complete data and experimental con-
dition as predictor, analysis 2 with data cleaned for
study protocol deviations and experimental condition
as predictor, and analysis 3 with cleaned data and
median CO
2
level as a predictor. In sensitivity analyses
for the primary outcome CP, first analyses 1–3 were
repeated after exclusion of observations from the
ÔusualÕcondition. Secondly, the effect of air quality
on CP was determined in analyses 1–3 including only
those observations in condition ÔbetterÕ, in which the
CO
2
level was <1000 ppm and those observations in
condition ÔworseÕin which CO
2
level was >2000 ppm.
All hierarchical models were implemented in the
software HLM 6.08.
Results
About half of the participating students were of grade
3 and about half were girls (Table 1). In seven percent
of the students, parents reported dyslexia.
Classrooms were between 59 and 71 m
2
of size. The
mean number of students of about 24 and the mean
room volume of 215 m
3
resulted in a mean air volume
of 9.2 m
3
available for each student (range, 6.9–
14.9 m
3
). Median room temperature on a test day
ranged between 20.0 and 26.3C (median = 23.6C,
10% percentile 22.1C, 90% percentile 25.1C) and was
similar on days with ÔusualÕ,ÔworseÕ,orÔbetterÕcondi-
tion. Median relative humidity in a classroom on a test
day ranged between 15.0% and 42.5% (med-
ian = 31.5%, 10% percentile = 22.7%, 90% percen-
tile = 39.4%) and was higher on days in ÔworseÕ
(median = 35.0%) than in ÔbetterÕcondition (med-
ian = 26.9%).
Figure 1 shows an example of the progression of
CO
2
level in one of the participating classrooms during
the six experimental days. The general pattern of CO
2
Table 1 Sociodemographic background of the children included in the RaBe study
Sequence of experimental condition
P-value
b
TotalUsual–better–worse Usual–worse–better
Grade
Grade 3 140 (55%) 64 (39%) <0.01 204 (49%)
Grade 4 114 (45%) 99 (61%) 213 (51%)
Sex
Girls 115 (45%) 86 (53%) 0.14 201 (48%)
Boys 139 (55%) 77 (47%) 216 (52%)
Place of birth (missing = 27)
a
Germany 224 (95%) 148 (96%) 0.58 372 (95%)
Other 12 (5%) 6 (4%) 18 (5%)
Dyslexia (missing = 29)
Yes 13 (6%) 13 (9%) 0.25 26 (7%)
No 222 (94%) 140 (92%) 362 (93%)
Hyperactivity (missing = 28)
Yes/do not know 27 (11%) 8 (5%) 0.03 35 (9%)
No 208 (89%) 146 (95%) 354 (91%)
Single parent (missing = 29)
Yes 32 (14%) 23 (15%) 0.70 55 (14%)
No 203 (86%) 130 (85%) 333 (86%)
Parental education (missing = 39)
Low 32 (14%) 21 (14%) 0.91 53 (14%)
Average/High 199 (86%) 126 (86%) 325 (86%)
Relative poverty
Yes 65 (26%) 33 (20%) 0.21 98 (24%)
No/unknown 189 (74%) 130 (80%) 319 (76%)
a
Missing refers to missing data because of incomplete fill-out of study questionnaires.
Subjects with missing data cannot be categorized.
b
Cochrane Mantel–Haenszel P-value of overall association to compare the distribution of
student characteristics in the two groups.
Fig. 1 Exemplary progression of CO
2
levels during the six
experimental days in one of the classrooms of the RaBe study
Effect of classroom air quality on studentsÕconcentration
381
level in schools, which is typical for classrooms relying
on natural ventilation, can be observed in our study
with classrooms relying on mechanical ventilation, too.
In the morning, CO
2
levels are low, increase during
school hours, decrease during breaks, and decrease
after students leave the school. In the experimental
condition, ÔusualÕ(blue lines) maximal CO
2
levels get
close to 1500 ppm, in the ÔbetterÕcondition (green lines)
CO
2
levels were decreased, and in the ÔworseÕcondition
(red lines) increased. The gray shade marks the relevant
time period from the beginning of the first school hour
until the end of the d2-test. Looking at all classrooms,
the fluctuation of CO
2
levels in the ÔbetterÕcondition
was smaller (average standard deviation of the mean
CO
2
levels = 141) than in the ÔworseÕcondition
(average standard deviation = 559). The distribution
of the minute-by-minute measurements in all class-
rooms is given in the Supporting information
(Table S1).
Table 2 shows the distribution of the median CO
2
values during this relevant time period for all 20
participating classes for the three experimental condi-
tions. It turned out to be difficult to regulate the
mechanical ventilation system in a way that ensured
the achievement of the planned CO
2
levels. Thus, only
on 20 of 40 days in condition ÔworseÕ, the median CO
2
level was between 2000 and 2500 ppm, and only on 22
of 40 days in condition ÔbetterÕ, the CO
2
level was
below 1000 ppm. However, on average, the median
CO
2
level in the ÔworseÕcondition (2115 ppm) was
1070 ppm higher than the CO
2
level in the ÔbetterÕ
condition (1045 ppm), which is significant
(P< 0.0001) in the t-test.
In each data collection, round between 394 and 401
of the 417 students returned the d2-test, of which 391–
397 could be used to calculate the test parameters,
resulting in overall 2366 values for each d2-parameter.
At the first occasion (round A), the mean values were
101 for CP, 270 for TN, and 13.7 for TE. During the
following occasions, the mean values of CP and TN,
but not TE, increased (Figure 2). Overall, values
ranged from )75 to 298 for CP, 102–657 for TN, and
0–324 for TE.
In the hierarchical linear model, we could not
observe a significant effect of the experimental condi-
tion on the primary parameter CP (Table 3). The CP
was decreased by 1.11 points at ÔworseÕin comparison
with ÔbetterÕair quality in analysis 1; however, this
difference was not statistically significant (95%
Fig. 2 Distribution of the test values during the six experimental
days (round A to F) by sequence of experimental conditions.
The boxplots show the median, the 25th and 75th percentile
(boundary of the box), the 10th and 90th percentile (whiskers)
and the 5th and 95th percentile (dots)
Table 2 Distribution of the median CO
2
levels in the 20 classrooms of the RaBe study by
experimental condition
Usual Worse Better
N(classrooms) 40 40 40
Mean 1371 2115 1045
<1000 ppm 5 1 22
1000–<1500 ppm 23 1 17
1500–<2000 ppm 7 14 –
2000–2500 ppm 5 18 1
>2500 ppm – 6 –
Twardella et al.
382
confidence interval CI )2.44 to 0.22). No significant
effect could be found, if observations were excluded
(analysis 2) or if the actual median CO
2
level instead of
the experimental condition was tested (analysis 3). In
sensitivity analysis after exclusion of observations from
round A and B, effect estimators were even lower and
also not statistical significant (analysis 1: 0.45, 95% CI
)1.96 to 1.07; analysis 2: 0.35, 95% CI )1.76 to 1.06;
analysis 3: 0.37, 95% CI )1.56 to 0.82). If only
observations in condition ÔbetterÕwere included, in
which the CO
2
level was <1000 ppm and only those
observations in condition ÔworseÕin which CO
2
level
was >2000 ppm, again, no significant effect of exper-
imental condition on CP was found (data not shown).
Similarly, no significant effect of experimental con-
dition or median CO
2
level on TN could be observed
(Table 3). With respect to TE, though, a significant
result was produced. In analysis 1, using all observa-
tions, the number of errors was increased by 1.34 (95%
CI )0.03 to 2.70) in the worse air-quality condition
compared to better air quality. If observations with
deviation from the protocol were excluded, the effect
became stronger and statistical significance was
reached. In analysis 2, the number of errors was
significantly increased by 1.65 (95% CI 0.42–2.87) in
case of worse air quality. Also, the TE was increased
with increased median CO
2
level. With an increase in
median CO
2
by 1000 ppm, TE increased by 1.19 (95%
CI 0.30–2.07).
Discussion
We could recruit 20 school classes in the RaBe study,
of which 417 students participated in data collection.
The regulation of the mechanical ventilation system to
reach the planned CO
2
levels in the classrooms turned
out to be challenging. However, a significant gradient
of 1070 ppm CO
2
between ÔworseÕand ÔbetterÕexper-
imental conditions could be achieved, indicating a
potentially relevant difference in air quality. The
hypothesis that worse air quality as indicated by
increased CO
2
level causes a reduction in CP in
students could not be confirmed with our study. While
the estimators for the effect air quality on CP did not
reach statistical significance, all point in the same
direction, that is, low performance at worse air quality.
Furthermore, results indicate a negative effect of worse
air quality on accuracy.
In the past, few studies have been conducted on the
CP of students and their relation to air quality as
indicated by CO
2
levels. In a review from the year 2005,
only one study was included which evaluated studentsÕ
performance depending on air quality (Mendell and
Heath, 2005). In this experimental study, a negative
effect of high CO
2
level on reaction time and perfor-
mance could be observed. Since then, seven other
studies on this issue have been published.
In an observational study from the USA, the air
quality with closed windows and active ventilation
system was measured in classroom of classes grade 5
during one school day and the ventilation rate
deduced. Standardized test scores based on math and
reading skills were obtained for the students. In the
pilot analysis of 50 school classes, the association
between the ventilation rate and test results was
significant only at P< 0.1 but not at P< 0.05
(Shaughnessy et al., 2006). In the main study, using
only data of those classes, in which ventilation rates
were below recommended guidelines (N= 87), a
significant association between the ventilation rate
and the results of the math and reading tests could
be observed (Haverinen-Shaughnessy et al., 2011). This
study provides indication of the relevance of indoor air
quality for long-term student achievement as indicated
by standardized tests. However, because of the obser-
vational design, the validity is limited.
The five remaining studies are experiments, in which
students were tested at different ventilation rates
resulting in different CO
2
levels. Naturally, only
short-term performance at the time of the specific air
quality can be measured in such a design. In Denmark,
two experimental studies have been undertaken in two
classrooms of one school with students of age ten to
twelve (Wargocki and Wyon, 2007a,b). Both were
cross-over trials in which CO
2
levels were regulated by
the ventilation system. However, additional window
opening was allowed. The performance of the students
was assessed by seven exercises exemplifying different
aspects of schoolwork. In the first experimental study,
the difference in CO
2
levels between study conditions
was low (on average 1270 with low and 920 ppm with
high ventilation), but still a significant higher speed of
work with high ventilation was found for five of the
Table 3 Estimators of the effect of the experimental condition (analyses 1 and 2) and the
actual CO
2
level (analysis 3) on the d2-test parameters with 95% confidence intervals
Number of observations in analysis
Reduction in test parameter (95% CI)
Analysis 1 Analysis 2 Analysis 3
Concentration performance
Worse compared to
better air quality
N= 2366
)1.11 ()2.44; 0.22)
N= 1962
)0.55 ()1.83; 0.73)
Median CO
2
(per
1000 ppm)
N= 1962
)0.76 ()1.86; 0.34)
Total number of characters
processed (TN)
Worse compared to
better air quality
N= 2366
)0.88 ()3.84; 2.08)
N= 2038
)0.11 ()3.23; 3.01)
Median CO
2
(per
1000 ppm)
N= 2038
)0.88 ()3.46; 1.70)
Total errors (TE)
Worse compared to
better air quality
N= 2366
1.34 ()0.03; 2.70)
N= 2254
1.65 (0.42; 2.87)
Median CO
2
(per
1000 ppm)
N= 2254
1.19 (0.30; 2.07)
Effect of classroom air quality on studentsÕconcentration
383
seven exercises (Wargocki and Wyon, 2007b). No effect
on accuracy was found. In the second study, because of
missing data, only four of the seven exercises could be
analyzed (Wargocki and Wyon, 2007a). For none of
those exercises, a significant influence of ventilation
was observed. The difference in CO
2
level in this trial
was even lower (about 775 vs. 1000 ppm).
In a study from England, computer-based tests were
conducted on 10 days in one class with students aged
ten to eleven years (Coley et al., 2007). Poor air quality
with mean CO
2
levels of 2909 ppm was reached by
restriction of window opening on 5 days, while on the
other 5 days by opening of the windows mean CO
2
levels of 690 ppm were achieved. For three of the 13
parameters of cognitive function, a significant influence
of air quality was observed. At good air quality,
reaction time was reduced and attention increased,
while at poor air-quality calmness was increased. It
remains unclear from the publication, whether the
learning effect was considered in the analysis. Further-
more, because study conditions were established by
window opening, students and teachers were not
blinded with respect to the study condition, which
could have an impact on test results.
In a second publication from the UK, preliminary
results of an experimental cross-over trial in one school
were reported (Bako-Biro et al., 2007). A direct air
supply system through the window was used to
establish two study conditions: (i) provide outdoor
air (mean CO
2
593–783 ppm); (ii) recirculate the
classroom air (mean CO
2
1638–4093 ppm). StudentsÕ
performance was tested with a 40-min paper-based test
(reading comprehension, addition, subtraction). No
effect of study condition on reading comprehension or
subtraction, but a significant improvement of addition
with provision of fresh air was observed.
Lastly, an experimental cross-over study in six
classes in Switzerland was conducted, and performance
assessed with the d2-test in students aged 15–16 years
(Ribic, 2008). Air quality was influenced by regulations
on window opening and CO
2
levels of 600–800 ppm for
good air quality and of at least 3000 ppm for low air
quality established. A significantly reduced CP was
observed at low air quality. However, students and
teachers were not blinded with respect to the study
condition. Furthermore, it remains unclear whether the
tests were completed under the same conditions, for
example, at the same time of day.
The above-mentioned studies mostly reported a
significant effect of air quality on at least some
parameters of studentsÕperformance. However, in
some studies, methodological limitations confine the
interpretation. The lack of blinding may have caused
false-positive associations. Parallel testing of multiple
performance parameters without correction of the P-
value weakens the interpretation of statistical signifi-
cance. Furthermore, the correlations between students
coming from the same class have not been accounted
for in the statistical analysis of most of the mentioned
studies. To neglect correlation between observations –
a clustered design – leads to an underestimation of
variance and thus can cause false statistical significance
(Donner and Klar, 2000). In our study, we tried to
avoid these methodological problems. We included a
relatively large number of classes and students and
were able to account for correlations between classes in
statistical analysis. Students, teachers, and test instruc-
tors were blinded with respect to the experimental
condition. We defined CP as our primary outcome to
avoid multiple testing. Results of secondary outcome
analysis are interpreted as explorative results.
With our RaBe study, we were not able to confirm
the results of former studies, which found a decreased
CP of children exposed to low air quality as indicated
by high CO
2
levels. The following limitations have to
be accounted for when interpreting our study results.
Mechanical ventilation in classrooms is very seldom in
Germany, and only recently, if schools are newly built
or if major renovations are undertaken, ventilation
systems are installed. Thus, although we aimed for 24
classes with 600 students, only 20 classes with 417
students took part. With a larger sample size, the effect
estimator might have become statistically significant.
However, even if statistically significant, it might be
questionable, whether a diminishment by 1% (effect
estimator 1.11 points, mean value in round A 101
points) would be regarded as a relevant change.
It was not always possible to adapt the timetable and
routine in schools to the optimal study design. Thus,
we could not avoid that in some cases classes had
physical education or left the classroom in the morning
before the test. We tried to compensate for these
deviations from study protocol by running analysis 2.
However, results did not change substantially. Still,
school is not a laboratory, and it is difficult to
standardize all possible influential factors.
Confounding is an unlikely explanation for the
observed results. First of all, one has to consider that
RaBe is a cross-over study, and thus, each student
serves as its own control. Thus, in the statistical model,
confounding of the effects size by any characteristic of
the student is not possible. However, a statistical
interaction cannot be excluded, which would be
interpreted as a differential effect of air quality
depending on student characteristics. As the model
that was used for RaBe was based on the assumption
that the effect of air quality on CP does not depend on
the characteristic of the students, it does not account
for interactions.
Secondly, RaBe was a randomized study. Random-
ization per se is a method to control confounding by
design (Rothman et al., 2008). If randomization is
successful, confounding by know as well as unknown
factors is prevented and thus adjustment in statistical
Twardella et al.
384
analysis not needed. However, the number of random-
ized units (classes) is relatively low and thus some
residual confounding possible.
To tackle the issue of residual confounding, the
following steps were taken: First of all, because of the
obvious strong enhancement of test parameters with
repeated testing, we assessed the effect of experimental
condition within a growth model, which captured this
change over time. Thus, all reported results are adjusted
by the Ôlearning effectÕ. Secondly, for the primary
outcome CP, we conducted adjusted analysis. Sociode-
mographic characteristics as well as room temperature
and relative humidity were assessed for relevance in the
growth model, and the effect of experimental condition
on CP was tested in an model adjusted for those factors
showing P< 0.1 in the growth model. In this model,
only the factor Ôsingle parentÕshowed a significant
interaction. While in the adjusted model there was a
significant reduction of CP in Ôlower air qualityÕ()1.85,
95% CI )3.38 to )0.33), in children with a single parent
this was changed to an increase by 2.83. There is no
obvious explanation for this interaction and chance
result because multiple testing is possible.
The CO
2
measurement range had an upper detection
limit of 3000 ppm, which was reached on eight of the
120 test occasions. The exceedance lasted for 1, 2, 3, 6,
7, 18, 39, or 49 min, respectively, on these eight
occasions. In none of the classes, the percentage of
the minute-by-minute CO
2
values on one test day,
which were above the detection limit reached 50%.
Thus, the calculation of median CO
2
levels was not
affected by this measurement limitation.
We were not able to achieve the exact CO
2
levels as
planned by regulation of the mechanical ventilation
system. However, first, we could show that on average
CO
2
levels were considerably higher in the ÔworseÕ
condition than in the ÔbetterÕcondition. Secondly, in
sensitivity analysis, we run a model in which only those
observations were included, in which the specifications
were achieved. Still in this analysis, no effect of
experimental condition on CP could be observed.
Thus, it seems implausible that the difficulties in
achieving specific CO
2
levels were responsible for the
absence of the effect.
Although students, teachers, and test instructors
were blinded with respect to the experimental condi-
tion, we cannot exclude that at least some subjects
could have become aware of the experimental condi-
tion during the test period because of body odors and
general stuffiness in the room. Thus, the reported
association might at least partly be due to insufficient
blinding.
The RaBe study differs with respect to the range of
observed CO
2
levels from other studies. Effects of CO
2
levels might only become evident when comparing
more distinct groups or air quality or achieving CO
2
levels far below 1000 ppm.
In RaBe, student performance was characterized
with the d2-test. In contrast to usual school examin-
ations, the test does not require specific skills but aims
to assess the aspect of concentration as a basic
prerequisite for the provision of any achievement
(Brickenkamp, 2002). In several studies, a positive
correlation between the CP derived from the d2-test
and intelligence and achievement motivation has been
observed (see for example Romainczyk, 2008 or
Schaal, 2004). It has also been shown that parameters
of the d2-test correlate with school marks (Brickenk-
amp, 2010). In some of the other published studies,
school work was used to characterize studentÕs perfor-
mance. While performance at school work is a more
direct measurement of a relevant outcome, which can
be obtained in a usual school situation, the knowledge
and skills of the students will impact the performance
to a larger extent particularly if the same examination
will be repeated at different times. Both studies using
tightly controlled tests as in our study (Coley et al.,
2007; Ribic, 2008) as well as studies using school
examinations to characterize student performance
(Wargocki and Wyon, 2007a) obtained a significance
influence of air quality. There are no indications for a
differential effect. The d2-test does assess concentration
only. Other possible effects of air quality such as an
increase in health symptoms and resulting absenteeism
are not covered.
In our study, air quality is described by carbon
dioxide concentration and not by ventilation rates. By
design, carbon dioxide levels were used to define
experimental groups. Carbon dioxide is a well-accepted
indicator of air quality, and parameters of ventilation
engineering are of less importance in a country such as
Germany with the majority of schools relying on
natural ventilation. Furthermore, as in RaBe class-
rooms serve as their own control and CO
2
generation
can be assumed constant within classrooms, differences
in ventilation rates in a classroom at different exper-
imental conditions will highly correlate with differences
in CO
2
levels. Even if ventilation rates were derived,
results of analyses 1 and 2 would not change because
the predictor variable was experimental group.
Using the median CO
2
levels, the situation in the
classrooms in RaBe can be easily compared to real-life
situation that is observed in classrooms in Germany as
has been described in the past (Fromme et al., 2008).
CO
2
is a well-known indicator of air quality and has
been related to health effects in the past (see for
example Shendell et al., 2004; Seppa
¨nen et al., 1999).
The reason for the absence of a significant effect on
the CP in our study does not become clear. We can
speculate, but not prove with data, that motivation of
students might have played a role. In RaBe, we
observed that the children looked forward to the test
and were highly motivated. The test was an interrup-
tion from the usual school work, and the children may
Effect of classroom air quality on studentsÕconcentration
385
have perceived it as a welcome diversion. In addition,
the d2-test lasts only about five minutes and thus
requires only short-term attention. It could be that for
a short period of time children can activate resources
even if exposed to low air quality. This would explain
why we could not observe an effect on the CP. It
cannot be excluded that in longer tests low air quality
eventually diminishes attention and CP.
Furthermore, while we could not observe an impact
on the primary outcome CP, our data do suggest a
decrease in the secondary outcome accuracy if children
are exposed to poor air quality. In comparison with the
mean total error of between 7.8 and 13.7, an increase in
1 to 1.5 points is an increase by about 10% and thus of
relevance. The association is significant if looking at
the effect of the experimental condition as well as
looking at the effect of the CO
2
level itself. In line with
the previous interpretation, one could argue that
children are able to achieve a normal CP for a short
time even if exposed to poor air quality, but while they
are able to sustain the processing speed they risk more
errors. Thus, tasks that require high precision might be
more strongly affected by poor air quality. However,
this result is in contradiction with the study of
Wargocki et al. in which increased ventilation in-
creased the speed of work but did not have any effect
on accuracy (Wargocki and Wyon, 2007b). The reason
for this contradictory result remains unclear.
Conclusion
We were not able to show a negative effect of air quality,
which was similar to that observed at wintertime in
classrooms relying on natural ventilation, on the CP of
students. Secondary analysis suggests an effect of air
quality on the processing accuracy. Our results are in
conflict with earlier studies, which suggested the pres-
ence of an effect on concentration. The causes of the
conflicting results remain unclear. It cannot be excluded
that significant results of earlier studies are at least partly
due to methodological limitations. While we tried to
avoid these limitations in RaBe, bias as a result of
varying circumstances in schools that are difficult to
standardize cannot be ruled out. One possible explana-
tion could be that poor air quality does not impair short-
term CP (as measured by the d2-test in our study) but
only shows its effect in longer concentration efforts.
Furthermore, air quality might be more relevant for the
precision than for the processing speed. Further research
is needed to clarify the relevance of indoor air quality on
long-term CP of students.
Acknowledgements
We wish to thank the participating schools, the
teachers, and students for their support of the study
as well as Ramona Grahle, Annemarie Hiergeist, Vera
Hoffmann, Petra Kaplan, Petra Panenka, Annette
Mangstl, Evelyn Schmidt, Ute Warmuth, and Simone
Zyzik-Zinn for their engagement in the data collection
and data management and Angelika Schwaiger for
assistance in organizational matters.
Competing interests
None.
Funding
The study was funded by the Deutsche Bundesstiftung
Umwelt (DBU), Az. 27549 – 25. The DBU was not
involved in the study design; in the collection, analysis,
and interpretation of the data; in the writing of the
report; or in the decision to submit the paper for
publication. The manuscript reflects the perception and
opinion of the contractor. It does not represent the
opinion of the DBU.
Supporting Information
Additional Supporting Information may be found in
the online version of the article:
Table S1 Distribution of the minute-by-minute mea-
surements of carbon dioxide concentration in each
classroom during the six 4-hour test periods.
Please note: Wiley-Blackwell are not responsible for
the content or functionality of any supporting materi-
als supplied by the authors. Any queries (other than
missing material) should be directed to the correspond-
ing author for the article.
References
Bako-Biro, Z., Kochhar, N., Clements-Cro-
ome, D.J., Awbi, H.B. and Williams, M.
(2007) Ventilation rates in schools and
learning performance. Proceedings of
Clima 2007 WellBeing Indoors, June 2007,
Helsinki.
Bates, M.E. and Lemay, E.P. Jr (2004) The
d2 Test of attention: construct validity
and extensions in scoring techniques,
J. Int. Neuropsychol. Soc.,10, 392–
400.
BMLFUW – Bundesministerium fu
¨r Land
und Forstwirtschaft, Umwelt und
Wasserwirtschaft (2006) Physikalische
Faktoren Kohlenstoffdioxid als
Lu
¨ftungsparameter, Wien.
Brickenkamp, R. (2002) Test d2.
Aufmerksamkeits- und Belastungstest.
Manual, 9th edn. Go
¨ttingen, Hogrefe
Verlag.
Brickenkamp, R., Schmidt-Atzert, L. and
Liepmann, D. (2010) Test d2 – Revision:
Aufmerksamkeits- und Konzentrationstest.
Go
¨ttingen, Hogrefe Verlag.
Bundesregierung (2005) Lebenslagen in
Deutschland – Zweiter Armuts- und
Twardella et al.
386
Reichtumsbericht.Ko
¨ln, Bundesanzeiger
Verlagsgesellschaft.
Coley, D.A., Greeves, R. and Saxby, B.K.
(2007) The effect of low ventilation rates
on the cognitive function of a primary
school class, Int. J. Vent.,6, 107–112.
Daisey, J.M., Angell, W.J. and Apte, M.G.
(2003) Indoor air quality, ventilation and
health symptoms in schools: an analysis
of existing information, Indoor Air,13,
53–64.
Donner, A. and Klar, N. (2000) Design and
Analysis of Cluster Randomization Trials
in Health Research, 1st edn. London,
Arnold.
Fromme, H., Heitmann, D., Dietrich, S.,
Schierl, R., Korner, W., Kiranoglu, M.,
Zapf, A. and Twardella, D. (2008) Rau-
mluftqualita
¨t in Schulen – Belastung von
Klassenra
¨umen mit Kohlendioxid (CO2),
flu
¨chtigen organischen Verbindungen
(VOC). Aldehyden, Endotoxinen und
Katzenallergenen [Air quality in schools -
classroom levels of carbon dioxide (CO2),
volatile organic compounds (VOC),
aldehydes, endotoxins and cat allergen],
Gesundheitswesen 70, 88–97.
Haverinen-Shaughnessy, U., Moschandreas,
D.J. and Shaughnessy, R.J. (2011) Asso-
ciation between substandard classroom
ventilation rates and studentsÕacademic
achievement, Indoor Air,21, 121–131.
Lahrz, T., Bischof, W. and Sagunski, H.
(2008) Gesundheitliche Bewertung von
Kohlendioxid in der Innenraumluft.
Mitteilungen der Ad-hoc-Arbeitsgruppe
Innenraumrichtwerte der Innenraumluf-
thygiene-Kommission des Umweltbunde-
samtes und der Obersten
Landesgesundheitsbeho
¨rden, Bundes-
gesundheitsbl Gesundheitsforsch
Gesundheitsschutz,51, 1358–1369.
Mendell, M.J. and Heath, G.A. (2005) Do
indoor pollutants and thermal conditions
in schools influence student performance?
A critical review of the literature, Indoor
Air,15, 27–52.
NO-Folkehelseinstituttet (1996). Anbefalte
faglige normer for inneklima. http://
www.fhi.no/dav/249C03CE-
C6614E87862368DA175E7A31.pdf.
Raudenbush, S.W. and Bryk, A.S. (2002)
Hierarchical Linear Models. Applications
and Data Analysis Methods, 2nd edn,
Thousand Oaks, London, New Delhi,
Sage Publications.
Ribic, W. (2008) Nachweis des Zusammen-
hangs zwischen Leistungsfa
¨higkeit und
Luftqualita
¨t, HLH Lu
¨ftung/Klima –
Heizung/Sanita
¨r – Geba
¨udetechnik, 59,
43–46.
Romainczyk, B. (2008) Bearbeitungsstrate-
gien in Konzentrationstests, Diploma
Thesis.
Rothman, K.J., Greenland, S. and Lash,
T.L. (2008) Modern Epidemiology, 3rd
edn, Philadelphia, Lippincott Williams &
Wilkins.
Schaal, S. (2004) Leistungsmotivation: eine
Untersuchung zur Validita
¨t verschiedener
Verfahren, Diploma Thesis, Marburg.
Seppa
¨nen, O.A., Fisk, W.J. and Mendell,
M.J. (1999) Association of ventilation
rates and CO
2
concentrations with health
and other responses in commercial and
institutional buildings, Indoor Air,9, 226–
252.
Shaughnessy, R.J., Haverinen-Shaughnessy,
U., Nevalainen, A. and Moschandreas,
D. (2006) A preliminary study on the
association between ventilation rates in
classrooms and student performance,
Indoor Air,16, 465–468.
Shendell, D.G., Prill, R., Fisk, W.J., Apte,
M.G., Blake, D. and Faulkner, D. (2004)
Associations between classroom CO
2
concentrations and student attendance in
Washington and Idaho, Indoor Air,14,
333–341.
UK Department of Education (2006) Venti-
lation of School Buildings. Building Bul-
letin 101. http://www.education.gov.uk/
schools/adminandfinance/schoolscapital/
buildingsanddesign/a0058229/ventilation-
and-indoor-air-quality-in-schools-
building-bulletin-101.
Wargocki, P. and Wyon, D.P. (2007a) The
effects of moderately raised classroom
temperatures and classroom ventilation
rate on the performance of schoolwork by
children (1257-RP), ASHRAE HVAC&R
Res.,13, 193–220.
Wargocki, P. and Wyon, D.P. (2007b) The
effects of outdoor air supply rate and
supply air filter condition in classrooms
on the performance of schoolwork by
children (1257-RP), ASHRAE HVAC&R
Res.,13, 165–191.
Effect of classroom air quality on studentsÕconcentration
387